English law does not need a bespoke AI liability framework to resolve civil claims for unintentional harm caused by artificial intelligence. That is the central finding of a legal statement published by the UK Jurisdiction Taskforce (UKJT) on September 14, 2026, which concludes that existing principles of contract and negligence are already well-equipped to handle these disputes - at least for now.
The UKJT, chaired by Sir Geoffrey Vos, Master of the Rolls, examined the circumstances in which English private law is likely to impose liability for AI harms. Its analysis carries significant weight even though the statement is not legally binding. The starting position is straightforward: AI itself cannot be sued because it lacks legal personality under English law. Liability must attach to a legal person - an individual or a corporate entity - and will only do so where that person has voluntarily assumed responsibility or where the law imposes it regardless.
The contractual backbone
Contract is the most common mechanism by which a party voluntarily takes on responsibility for risk. Where a contract governs the situation in which AI harm arises, liability becomes a matter of interpreting the relevant contractual provisions. The UKJT considers that in many - perhaps most - cases, the chain of contracts between actors in the AI supply and use chain will be the most important legal framework. Applying the established rules of contractual interpretation to these cases is relatively straightforward.
The legal statement therefore focuses its attention on noncontractual cases, where liability will largely be determined by the law of negligence. This is where the analysis gets more granular, examining each element of a negligence claim through an AI-specific lens.
Duty of care in the AI supply chain
Many circumstances in which a duty of care is owed are already well established - manufacturers to consumers, doctors to patients, employers to employees. The UKJT's view is that adding AI to a scenario simply treats the technology as a tool of those who exercise control over it; it does not change the underlying position on whether a duty exists. In novel situations, courts will look to the closest analogies in existing law and apply them by incremental extension.
How far up or down the AI supply chain a duty of care extends depends primarily on whether the fault and the harm were foreseeable. A foundation model developer would be unlikely to owe a duty for harms arising from unforeseeable uses of its model by a downstream participant who failed to conduct sufficient testing. But that same developer may owe a duty where harm is foreseeably suffered as a result of foreseeable use. The UKJT also flags that a party in the supply chain that has taken active steps to push out software updates to address risks of harm may be more likely to be deemed to owe a duty of care.
Standard of care and professional negligence
Once a duty is established, the standard of care in nonprofessional cases is to act as a reasonable person would in the circumstances. This is highly fact-specific and will typically require expert evidence. At a minimum, anyone in the AI supply chain should be aware of and implementing relevant industry guidance. The UKJT points to the AI Standards Hub as a "useful yardstick" for what represents reasonable practice.
For professionals - lawyers, accountants, doctors - the standard is higher. They must act as a reasonable member of their profession with comparative rank and specialisation would act. This includes exercising reasonable care and skill in deciding whether and how to use AI. The UKJT lists several indicators of breach, including failure to conduct proper due diligence on an AI system, insufficient understanding of the system being deployed, failure to ensure adequate testing, and failure to exercise oversight of the system's output.
The standard is not static. "Given the rampant pace of development in AI systems, what is reasonable now may be negligent in the not-too-distant future," the statement warns. Where a professional cannot practically undertake due diligence or monitor outputs, a court may conclude either that there is a limit to what can reasonably be expected, or that the professional should not have used the AI tool at all. Conversely, failing to use AI when a reasonable professional would have done so can also lead to liability.
Causation: the battleground issue
Causation is likely to be one of the most contentious elements of AI harm claims. The test in negligence is the "but for" standard: the loss would not have been suffered but for the breach. The UKJT highlights two significant sources of causal uncertainty - gaps in evidence because material has been destroyed or not gathered, and the opacity of AI systems themselves.
To address evidential gaps, the UKJT suggests courts might take a "benevolent" approach to affected third parties where a defendant failed to record information it reasonably should have. Expert evidence obtained through experimentation, such as running simulations, may also fill gaps. English civil law operates on the balance of probabilities, not certainty, which the UKJT says "is well suited for experimental proofs."
On AI opacity, the statement points to two niche principles from discrete areas of negligence law that could bridge causal gaps: the "material increase in risk" test and the "material contribution" test. Whether courts would extend these principles beyond their existing narrow categories remains uncertain.
Legal causation - whether something has intervened to break the chain of causation - raises questions about deliberate misuse and AI autonomy. On deliberate misuse, a party in the AI supply chain might be held responsible for creating a source of danger by bringing an AI system into existence without suitable safeguards, but the UKJT says such cases "are likely to be rare." On autonomy, courts will be "very slow" to find that an AI system learning new behaviours constitutes a novel intervening act that breaks the chain of causation. Policy considerations favour finding a legal person liable where harm has been caused, especially if that person could benefit from the activity that gave rise to the harm.
Strict liability and the Consumer Protection Act
Outside of negligence, the general position in English law is that the risk of loss for nondeliberate harm lies where it falls. The major exception is where an AI system is incorporated into a tangible product, triggering the Consumer Protection Act 1987, which imposes strict no-fault liability where a product is defective. The Law Commission has announced an intention to review the status of "pure software" under the Act, so AI systems may fall within its ambit in the future.
Why this matters for legal professionals
The UKJT's analysis confirms that existing legal frameworks are being stretched to accommodate AI, but the stretch points are clear. Causation will be the hardest element to prove, and the standard of care for professionals using AI is a moving target that demands continuous attention. The most immediate action points are to control risk exposure through contracts, map the potential scope of your duty of care, maintain a system for tracking and implementing industry guidance, and keep all of this current as both the technology and its regulation evolve. The statement also flags contributory negligence as a live risk: a commercial user who relies on AI output without verification where the consequences of error are serious should expect reduced damages.
Your membership also unlocks:
